About
Accomplished Executive Leader with Expertise in Driving Business Growth through Strategic…
Services
Articles by ✔Michael
Activity
3K followers
-
✔Michael Couvillion reposted this✔Michael Couvillion reposted thisAt $10M–$30M, the best AI opportunities usually aren’t AI problems. They’re expensive business problems. A few examples from Friday’s Agentic Society boardroom: -- A team was manually pulling reports from software with no API. Now an agent logs in overnight and has the data waiting for procurement in the morning. -- A key employee left and years of institutional knowledge walked out with them. They’re now building a company brain so it doesn't happen again. -- A $6M founder is automating analysis and client delivery so his advisory firm can grow without adding headcount at the same rate. -- Another is connecting website behavior, session recordings, and conversion data to find exactly where qualified buyers drop off and what to fix. -- Another built a system across 5 inboxes, calendars, and Slack so only the decisions that actually need him reach him. This is the filter we relentlessly apply: -- Where are you paying expensive people to do repetitive work? -- Where does critical knowledge live in one person’s head? -- Where is the founder still the bottleneck? -- Where is revenue leaking? Bring that problem into the room. We’ll attack it from growth, ops, AI architecture and implementation - then figure out what can be automated, delegated, or rebuilt entirely. PS - We're sold out for the mastermind this Friday. Founders are doing anywhere from $1M-$50M+ annually. Stoked to meet some incredible builders and see what problems they bring into the room.
-
✔Michael Couvillion posted thisToday's LinkedIn rant. Got a new phone. Went to open the LinkedIn mobile app and I have to sign in again. 1) It says it sent a confirmation to my mobile device. My old one. That I no longer own and had factory reset. 2) Try the SMS. Get the ever so helpful "something went wrong" error message that if any dev of mine ever enters such an error message they will be immediately terminated for being a lazy hack instead of a professional software engineer. 3) Can log into the LinkedIn website. Attempt to see if putting another phone number will help. Sure, first authenitcate to your old phone you no longer have. 4) See step 1. So, to sign into the app I have to confirm on a device I no longer have. Oh, they have that option "I can't access that device now" so what do they do? Pop up a QR code to SCAN WITH THE PHONE YOU NO LONGER HAVE. They should have let AI write this logic loop. The human produced slop code is often 10 times worse. LinkedIn, where Catch-22 login logic brought to you by experts.
-
✔Michael Couvillion shared thisFun times fixing a problem where I was at fault and had to have myself humbled.I was doing an AI Test Suite wrong. Learn from my mistake.I was doing an AI Test Suite wrong. Learn from my mistake.
-
✔Michael Couvillion posted thisI was using Claude Code wrong. And it was costing me hundreds of thousands of tokens if not millions. That's real dollars. How was I doing it wrong? One, by just being a bit lazy. Launch a terminal, fire up claude, default to Opus and start working. Sometimes kick up to Fable for a planning session then back to Opus for coding, tickets etc. It wasn't until I realized I had closed 3 documentation update tickets using Opus. Burned about 200K of tokens doing that. For about 10 lines of texts in 3 files. For grins I switched the model to Haiku and asked it to do the same thing. Exact same changes made and cost about 10K tokens. About 5% of my tickets were that way. Rookie mistake. I'm using a sledge hammer to drive staples. First, built a simple deterministic triage bot. Doc updates don't even go to Claude. Shunt that down to my hermes agent and let qwen3.8 handle the simple doc edits. No tokens burned. Next, I fed the git log of all 11 of my current projects (13 repos in total and ~ million LOC overall) to Fable to analyze which tickets were far simpler to implement and didn't need Opus or Fable. Turns out about 45% of the tickets and PRs (almost all were code) could have easily been done by Sonnet 5. The specs and/or tickets were well documented and had sufficient unti tests and data evals that Sonnet 5 didn't really need a lot of thinking. About 15% were edits to my dark factory specs. Those are complex and require a lot of context. Fable handles those. Added that to the determinsitic triage engine. Re-wrote the dispatcher to simpy use Sonnet 5 first on a ticket and if there are bugs/mistakes made then it automatically bumps to Opus for a review/rewrite. A handoff if you will. There is some cost there but the end result was good. I burned this week's tokens over the weekend running a big test. I took a sample set of 1000 PRs and their associated commits and ran them the old way and the with the new triage/escalation process. Results were the code written was nearly identical to the "all Opus all the time" to "divvy it up properly". All Opus burned ~300% more tokens. I am going to be using this new orchestrator. Especially as I am now going to build out a development manager agent. It won't write code but it will orchestrate ticket assignment using this new method, which is probably an old method for others. Downside is I now have to wait until Wed to use Claude, but that's ok because I just picked up an AMD Strix Halo 128GB machine to setup.
-
✔Michael Couvillion posted thisI cancelled about $200 of monthly software subscriptions today. About $50 was just house cleaning of things I no longer use. About $50 was dumping a paid subscription for something that there was an open source and free thing for. The last $100 was for apps that I did use but only about 30% of their capability. What I did instead was burn a Sunday's worth of tokens with Claude code to build a personal app just for me. I am the only user and I am going to be the only user. I am seeing this happen on a larger scale too. I belong to a group of business owners who are looking for pragmatic solutions to their everyday problems. They are not technologists or "AI Engineers". Most are in the trades and they are using AI tools to automate those things that eat an hour or two of their day or week. The result is they are cancelling expensive SaaS products for their own personal software. Yes, they are taking on the burden of maintaining it and there is always a "build vs buy" decision process but more and more the answer is "build" because it can be done a lot easier. This week alone I saw a manufacturing warehouse cancel a $30K/year SaaS product because their homegrown version had the features they waited years for and didn't have a ton of features they didn't like. They understood the problem and building the app wasn't a barrier. Their expertise in how it should work is the barrier. I've seen an HVAC company build their own scheduling software that reduced the time on a job enough that they could expand the number of customers they could serve by simply routing the service trucks smarter. I'm not sure it will fully hit Enterprise levels, certainly to some extent, but in the small to medium business model the concept of "personal apps" is starting to take hold and have a significant impact. Writing code is no longer a barrier to entry for the SMB market.
-
✔Michael Couvillion shared thishttps://lnkd.in/gPf5cKvR Had the pleasure of working with Zachary M. Schwitzky, Caroline Tarpey and 👩🏻💻Erica Byrd as an investorm advisor and consultant for Limbik for the last three years and happy to their success. Another succesful exit!
-
✔Michael Couvillion reposted thisCongratulations to Limbik on being acquired by Burson. I joined Limbik as CRO/COO in 2022, inspired by CEO and Founder Zachary M. Schwitzky, helping build the GTM engine and organizational foundation. I later stayed on as an Operating Advisor. Along the way I worked closely and occasionally shared the stage with the Burson team on AI and reputation protection including the visionary Chad Latz - so this outcome feels full circle. Proud to have worked alongside the devoted founding team Zachary M. Schwitzky, Josh Levin, and Milhem Calderon - congratulations - this moment is yours to savor. Thank you to my friend and partner ✔Michael Couvillion for believing in what we were building enough to come on the journey as an advisor and investor. And finally hats off to 👩🏻💻Erica Byrd who was instrumental in transforming the platform to what it is today. Congrats to the whole team - excited for what's next with Burson.✔Michael Couvillion reposted thisBurson acquires Limbik, the cognitive-AI company and co-creator of Decipher. This acquisition brings Limbik's proprietary AI models, predictive engines and engineering fully in-house, allowing us to accelerate development across our entire innovation suite and expand Decipher's reach across our network. Read the full announcement: https://lnkd.in/etpmM5rE
-
✔Michael Couvillion posted thisAccording to my trainer drinking 5 cups of coffee and running my heart rate to 150bpm does not count as cardio. I disagree. Launching a new business can do the same thing. I mean, if you aren't having at least 2 issues a day running your heart rate up are you even an entrepreneur?
-
✔Michael Couvillion posted thisRevisited using ollama and a free model. In my past experiments I had 0% success rate in getting any free model to produce any code that would be usable. I thought all the claims of "I wrote a whole app on my mac book while flying to Europe and it was done in 6 hours" were just lies. So, thinking something was off I decided to try again. So, I tried it again using a 28GB model that just fit on the two GPUs I had in the server box. I gave it 8 regular bug tickets off one my non-critical apps. I had claude be the judge of the results. At first all failed, then I made some changes. High level score. 4 of the bugs were fixed and the code was good. The other 4 turned out to need some architectural decisions and were tagged as such. Even claude said they were more complex and would likely need Opus 5 or Fable to implement. So what was the difference? One was to turn off thinking. qwen3.6 is a thinking model and it would simply use up all 32K of tokens thinking and return nothing. I'm not sure why it was growing to 32K when everywhere else it never got above 6000. Anyhow, turned off thinking and I had to tune the temp to 0.15 (from 1.0) and the presence_penalty from 1.5 to 0. That presence penalty is natively destructive to coding so not sure why it is the default. I haven't done further testing with other models yet. It was trying to push e to use other cloud models but since they are the same price as a basic claude or codex license I didn't see the point.
✔Michael Couvillion commented on a post
1d
Just prior to COVID I was only coming into the office twice a week for this exact reason. Half the team I managed was in another city and the other half in a different country. The 3 hours a day of being stuck in traffic to just be sat in the little dinky conference room with no window to be on a zoom call with them was a waste of time, energy and my soul. I was fortunate to have a boss at that time that understood it and measured the output I produced rather than focusing on how much time I sat in a chair in a dark room.
✔Michael Couvillion replied to a comment
3d
Brian Grim Better to use AI to build deterministic tools to do the reviews because LLMs can be interpretive and not always as consistent as you need it to be.
-
✔Michael Couvillion liked this✔Michael Couvillion liked thisAt $10M–$30M, the best AI opportunities usually aren’t AI problems. They’re expensive business problems. A few examples from Friday’s Agentic Society boardroom: -- A team was manually pulling reports from software with no API. Now an agent logs in overnight and has the data waiting for procurement in the morning. -- A key employee left and years of institutional knowledge walked out with them. They’re now building a company brain so it doesn't happen again. -- A $6M founder is automating analysis and client delivery so his advisory firm can grow without adding headcount at the same rate. -- Another is connecting website behavior, session recordings, and conversion data to find exactly where qualified buyers drop off and what to fix. -- Another built a system across 5 inboxes, calendars, and Slack so only the decisions that actually need him reach him. This is the filter we relentlessly apply: -- Where are you paying expensive people to do repetitive work? -- Where does critical knowledge live in one person’s head? -- Where is the founder still the bottleneck? -- Where is revenue leaking? Bring that problem into the room. We’ll attack it from growth, ops, AI architecture and implementation - then figure out what can be automated, delegated, or rebuilt entirely. PS - We're sold out for the mastermind this Friday. Founders are doing anywhere from $1M-$50M+ annually. Stoked to meet some incredible builders and see what problems they bring into the room.
-
✔Michael Couvillion liked this✔Michael Couvillion liked thisThe picture place to soak in the final stretch of summer. Today, we revisit this serene and layered home facing Lake Washington by Katie LeClercq. Rediscover the full home tour featured in one of our past Pacific Northwest issues at the link in bio. Architecture: Katie LeClercq Design Studio and Okano Picard Studio; Interior Design: Studio LeClercq; Home Builder: Mercer Builders LLC; Photographer: Aaron Leitz
-
✔Michael Couvillion liked this✔Michael Couvillion liked thisI present to you a FIVE HOUR pod cast. Yep…and if you are a practicing software CTO you need to invest the time. This is gold from a GOAT. https://lnkd.in/gakhiMQrDHH: Future of Programming, AI, Agentic Engineering, Vibe Coding & Linux | Lex Fridman Podcast #501DHH: Future of Programming, AI, Agentic Engineering, Vibe Coding & Linux | Lex Fridman Podcast #501
Experience & Education
Volunteer Experience
-
Flight Crew
Commemorative Air Force, CENTEX Wing
- 7 years 5 months
Education
-
Coach/Mentor
51Vets / Honor Foundation / EliteMeetUS
- Present 6 years 9 months
i18n_veterans
I coach military veterans as they transition from military life to civilian life. Everything from resume writing, business advice and how to succeed in the new world they are entering.
Publications
-
Enterprises Embrace Obama’s HPC Moon Shot
EnterpriseTech
See publicationI wasn't the author but the author quoted me in the article.
Projects
-
Transition Coaching
Working with two elite forces veterans (former SEAL and a former fighter pilot) as they transition from the military life to civilian life. Coaching them on how to adapt their abilities into assets for any company (even their own) and deal with the culture shock of their new world.
-
Agentic AI SDLC Practice
-
I am building an Agentic AI SDLC process and tools for the new world of software development.
Honors & Awards
-
2022 Brandon Hall Group HR Technology Excellence Award
Brandon Hall Group
2022 Brandon Hall Group HR Technology Excellence Award – Bronze for Best Advance in Recognition and Rewards Technology
-
2019 Distinguished Alumni, La. Tech University College of Engineering and Science
Louisiana Tech University, College of Engineering and Science
-
5 CTOs to Know in Austin
Built In Austin
https://www.builtinaustin.com/2015/11/16/ctos-you-should-know-mike-couvillion-drillinginfo
Organizations
-
Agentic Society
Founder/VIP
- PresentBuild AI Agents & Scale Profits, Together. You're running a 7-8 figure business and AI keeps moving faster than your team does. Some people are using it, most aren't, and nobody has a clear system yet. That changes when you're in the right room. Agentic Society is where Austin's savvy business owners come together, build impactful AI agents, and lead their teams into the Agentic Workforce.
Recommendations received
21 people have recommended ✔Michael
Join now to viewOther similar profiles
Explore more posts
-
Dr. Denise Dennis
Are your business texts putting you at risk for $5,000+ fines? Texas SB140 just changed how EVERY coach sends SMS. Are you ready for the fallout? On this episode of the Automated Coach Podcast, Chief Automation Officer Dr. Denise Dennis breaks down the brand-new law that impacts anyone texting Texas residents (yes, even if you’re not in Texas!). Find out the 5 things you MUST do to keep your coaching business safe while still growing your audience before it’s too late! Tune in now for practical steps, real talk, and smart strategies you can implement today. Listen to the latest podcast episode here: https://lnkd.in/g8_dCitX . . . #AutomatedCoachPodcast #businessautomation #SMSmarketing #coachingsuccess #telemarketing #legalcompliance #marketingautomation #emailmarketing
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content